munibz/pii-distilbert-en

PII Detection — DistilBERT (English)

Fine-tuned distilbert-base-cased for PII (Personally Identifiable Information) detection on the English subset of ai4privacy/pii-masking-300k.

Results

• F1: 0.96
• 30k training examples, 5k validation, 6 epochs on a Colab T4

Entity types

PERSON, EMAIL, PHONE, USERNAME, ID_NUMBER, ADDRESS, IP_ADDRESS, URL, DATE_TIME, PASSWORD, DEMOGRAPHIC, OTHER

Usage

from transformers import pipeline

ner = pipeline("ner", model="munibz/pii-distilbert-en", aggregation_strategy="simple")

text = "Hi, I'm John Smith. Email me at john.smith@gmail.com."
print(ner(text))

Training

• Base model: distilbert-base-cased
• Max length: 384
• Batch size: 32
• Learning rate: 3e-5 (cosine schedule, 10% warmup)
• Epochs: 6 (early stopping patience 2)
• FP16 on T4 GPU